Self-Driving Car Engineer Nanodegree

Deep Learning

Project: Build a Traffic Sign Recognition Classifier

In this notebook, a template is provided for you to implement your functionality in stages which is required to successfully complete this project. If additional code is required that cannot be included in the notebook, be sure that the Python code is successfully imported and included in your submission, if necessary. Sections that begin with 'Implementation' in the header indicate where you should begin your implementation for your project. Note that some sections of implementation are optional, and will be marked with 'Optional' in the header.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. In addition, Markdown cells can be edited by typically double-clicking the cell to enter edit mode.

Step 0: Load The Data

In [342]:
# Load pickled data
import pickle

# TODO: Fill this in based on where you saved the training and testing data

training_file = 'traffic-signs-data/train.p'
testing_file = 'traffic-signs-data/test.p'

with open(training_file, mode='rb') as f:
    train = pickle.load(f)
with open(testing_file, mode='rb') as f:
    test = pickle.load(f)
    
X_train, y_train = train['features'], train['labels']
X_test, y_test = test['features'], test['labels']

Two lines below disable scrolling of the output cell

In [343]:
%%javascript
IPython.OutputArea.auto_scroll_threshold = 9999;

Step 1: Dataset Summary & Exploration

The pickled data is a dictionary with 4 key/value pairs:

  • 'features' is a 4D array containing raw pixel data of the traffic sign images, (num examples, width, height, channels).
  • 'labels' is a 2D array containing the label/class id of the traffic sign. The file signnames.csv contains id -> name mappings for each id.
  • 'sizes' is a list containing tuples, (width, height) representing the the original width and height the image.
  • 'coords' is a list containing tuples, (x1, y1, x2, y2) representing coordinates of a bounding box around the sign in the image. THESE COORDINATES ASSUME THE ORIGINAL IMAGE. THE PICKLED DATA CONTAINS RESIZED VERSIONS (32 by 32) OF THESE IMAGES

Complete the basic data summary below.

In [344]:
### Replace each question mark with the appropriate value.
import numpy as np

# TODO: Number of training examples
n_train = len(X_train)

# TODO: Number of testing examples.
n_test = len(y_train)

# TODO: What's the shape of an traffic sign image?
image_shape = X_train.shape[1:3]

# TODO: How many unique classes/labels there are in the dataset.
n_classes = len(np.unique(y_train))

print("Number of training examples =", n_train)
print("Number of testing examples =", n_test)
print("Image data shape =", image_shape)
print("Number of classes =", n_classes)

Number of training examples = 39209
Number of testing examples = 12630
Image data shape = (32, 32)
Number of classes = 43

Visualize the German Traffic Signs Dataset using the pickled file(s). This is open ended, suggestions include: plotting traffic sign images, plotting the count of each sign, etc.

The Matplotlib examples and gallery pages are a great resource for doing visualizations in Python.

NOTE: It's recommended you start with something simple first. If you wish to do more, come back to it after you've completed the rest of the sections.

In [345]:
def signname(index):
    import pandas as pd
    signnames = pd.read_csv('signnames.csv')
    return signnames[signnames.ClassId == index].SignName.tolist()[0]
In [346]:
### Feel free to use as many code cells as needed.
import matplotlib.pyplot as plt
import random
# Visualizations will be shown in the notebook.
%matplotlib inline

f = plt.figure(figsize=(20,10))
for i in range(8):
    ax = f.add_subplot(2, 4, i+1)
    index = random.randint(0, len(X_train))
    image = X_train[index].squeeze()
    plt.imshow(image, cmap="gray")
    ax.set_title(signname(y_train[index]))

Step 2: Design and Test a Model Architecture

Design and implement a deep learning model that learns to recognize traffic signs. Train and test your model on the German Traffic Sign Dataset.

There are various aspects to consider when thinking about this problem:

  • Neural network architecture
  • Play around preprocessing techniques (normalization, rgb to grayscale, etc)
  • Number of examples per label (some have more than others).
  • Generate fake data.

Here is an example of a published baseline model on this problem. It's not required to be familiar with the approach used in the paper but, it's good practice to try to read papers like these.

NOTE: The LeNet-5 implementation shown in the classroom at the end of the CNN lesson is a solid starting point. You'll have to change the number of classes and possibly the preprocessing, but aside from that it's plug and play!

Implementation

Use the code cell (or multiple code cells, if necessary) to implement the first step of your project. Once you have completed your implementation and are satisfied with the results, be sure to thoroughly answer the questions that follow.

In [347]:
### Preprocess the data here.
### Feel free to use as many code cells as needed.
import cv2
def grayscale(img):
    """Applies the Grayscale transform
    This will return an image with only one color channel
    but NOTE: to see the returned image as grayscale
    you should call plt.imshow(gray, cmap='gray')"""
    return cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
def data_pp(X, gray=0, norm=0):
    X_pp = np.empty([len(X),32,32,1])
    # transform input to Gray
    if (gray==1):
        X_pp = np.empty([len(X),32,32,1])
        for i in range(len(X)):
            X_pp[i] = np.reshape(grayscale(X[i]), (32,32,1))
    else:
        X_pp = np.copy(X)
    #normalize input (X - 128)/128
    if (norm==1):
        X_pp = (X_pp - 128)/128
    return X_pp
In [348]:
X_train_pp = data_pp(X_train, gray=1, norm=1)
X_test_pp = data_pp(X_test, gray=1, norm=1)

Question 1

Describe how you preprocessed the data. Why did you choose that technique?

Answer:

  • transformation into grayscale
  • zero-center and normalize input pixels
In [350]:
### Generate data additional data (OPTIONAL!)
### and split the data into training/validation/testing sets here.
### Feel free to use as many code cells as needed.

# splitting train data into train and validate sets with ratio 95:5
from sklearn.model_selection import train_test_split
X_train_pp, X_validation, y_train, y_validation = train_test_split(
    X_train_pp, y_train, test_size=0.05, random_state=42)

Question 2

Describe how you set up the training, validation and testing data for your model. Optional: If you generated additional data, how did you generate the data? Why did you generate the data? What are the differences in the new dataset (with generated data) from the original dataset?

Answer:
For this project training and testing data set was provided with 39209 and 12630 examples respectively. In order to prevent overfitting, the training data set was split into training and validation sets with ration 90:10.

Data augmentation (to be added)

In order to improve the accuracy of the convolutional neural network, the artificial increase of the data set or data augmentation might be used. The increase of the data set might be done by generation new images from the transformation of the given:

  • horizontally flipping;
  • random crops and color jittering;
  • rotation;
  • random scaling;
  • changing saturation and value (S and V components of the HSV color space) of all pixels to a power between 0.25 and 4 (same for all pixels within a patch), multiply these values by a factor between 0.7 and 1.4, and add to them a value between -0.1 and 0.1
  • adding a value between [-0.1, 0.1] to the hue (H component of HSV) of all pixels in the image/patch; etc.[1]
In [351]:
### Define your architecture here.
### Feel free to use as many code cells as needed.
In [352]:
#LeNet
import tensorflow as tf
from sklearn.utils import shuffle

EPOCHS = 10
BATCH_SIZE = 128
input_channels = len(X_train_pp[0][0][0])

from tensorflow.contrib.layers import flatten

mu = 0
sigma = 0.1
    
weights = {
    'layer_1': tf.Variable(tf.truncated_normal(
        [5, 5, input_channels, 6])),
    'layer_2': tf.Variable(tf.truncated_normal(
        [5, 5, 6, 16])),
    'fully_connected1': tf.Variable(tf.truncated_normal(
            shape=(400, 120), mean = mu, stddev = sigma)),
    'fully_connected2': tf.Variable(tf.truncated_normal(
            shape=(120, 84), mean = mu, stddev = sigma)),
    'fully_connected3': tf.Variable(tf.truncated_normal(
            shape=(84, 43), mean = mu, stddev = sigma))
}
biases = {
    'layer_1': tf.Variable(tf.zeros(6)),
    'layer_2': tf.Variable(tf.zeros(16)),
    'fully_connected1': tf.Variable(tf.zeros(120)),
    'fully_connected2': tf.Variable(tf.zeros(84)),
    'fully_connected3': tf.Variable(tf.zeros(43))
}
def LeNet(x):    
    # Hyperparameters
    
    padding = 'VALID'
    # TODO: Layer 1: Convolutional. Input = 32x32xinput_channels. Output = 28x28x6.
    conv1 = tf.nn.conv2d(x, weights['layer_1'], strides=[1,1,1,1], padding='VALID') + biases['layer_1']

    # TODO: Activation.
    conv1 = tf.nn.relu(conv1)

    # TODO: Pooling. Input = 28x28x6. Output = 14x14x6.
    conv1 = tf.nn.max_pool(conv1, ksize=[1,2,2,1], strides=[1,2,2,1], padding='VALID')
    
    # TODO: Layer 2: Convolutional. Output = 10x10x16.
    conv2 = tf.nn.conv2d(conv1, weights['layer_2'], strides=[1,1,1,1], padding='VALID') + biases['layer_2']
    
    # TODO: Activation.
    conv2 = tf.nn.relu(conv2)

    # TODO: Pooling. Input = 10x10x16. Output = 5x5x16.
    conv2 = tf.nn.max_pool(conv2, ksize=[1,2,2,1], strides=[1,2,2,1], padding='VALID')

    # TODO: Flatten. Input = 5x5x16. Output = 400.
    flatten = tf.contrib.layers.flatten(conv2)

    # TODO: Layer 3: Fully Connected. Input = 400. Output = 120.
    fc3 = tf.add(tf.matmul(flatten, weights['fully_connected1']),biases['fully_connected1'] )
    
    # TODO: Activation.
    fc3 = tf.nn.relu(fc3)

    # TODO: Layer 4: Fully Connected. Input = 120. Output = 84.
    fc4 = tf.add(tf.matmul(fc3, weights['fully_connected2']), biases['fully_connected2'])
    
    # TODO: Activation.
    fc4 = tf.nn.relu(fc4)

    # TODO: Layer 5: Fully Connected. Input = 84. Output = 43.
    fc5 = tf.add(tf.matmul(fc4, weights['fully_connected3']), biases['fully_connected3'])
    logits = fc5
    
    return logits

#Features and labels
x = tf.placeholder(tf.float32, (None, 32, 32, input_channels))
y = tf.placeholder(tf.int32, (None))
one_hot_y = tf.one_hot(y, 43)

# Training Pipeline
rate = 0.001

logits = LeNet(x)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits, one_hot_y)
loss_operation = tf.reduce_mean(cross_entropy)
optimizer = tf.train.AdamOptimizer(learning_rate = rate)
training_operation = optimizer.minimize(loss_operation)

# Model Evaluation
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(one_hot_y, 1))
accuracy_operation = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
saver = tf.train.Saver()

def evaluate(X_data, y_data):
    num_examples = len(X_data)
    total_accuracy = 0
    sess = tf.get_default_session()
    for offset in range(0, num_examples, BATCH_SIZE):
        batch_x, batch_y = X_data[offset:offset+BATCH_SIZE], y_data[offset:offset+BATCH_SIZE]
        accuracy = sess.run(accuracy_operation, feed_dict={x: batch_x, y: batch_y})
        total_accuracy += (accuracy * len(batch_x))
    return total_accuracy / num_examples

LeNet

  1. RGB -> GRAY
    normalized
    Train - Validation: 0.1
    Validation Accuracy = 0.946
    Test Accuracy = 0.853
  2. RGB
    normalized
    input 32x32x3
    Validation Accuracy = 0.223
    Test Accuracy = 0.196
  3. RGB
    not normalized
    input 32x32x3
    Validation Accuracy = 0.052
    Test Accuracy = 0.059
  4. GRAY
    normalized
    Train - validation: 70:30
    Validation Accuracy = 0.913
    Test Accuracy = 0.799
  5. GRAY
    normalized
    Train - validation: 80:20
    Validation Accuracy = 0.945
    Test Accuracy = 0.841
  6. GRAY
    normalized
    Train - validation: 95:5
    Validation Accuracy = 0.942
    Test Accuracy = 0.853

Question 3

What does your final architecture look like? (Type of model, layers, sizes, connectivity, etc.) For reference on how to build a deep neural network using TensorFlow, see Deep Neural Network in TensorFlow from the classroom.

Answer: First arcitecture I've tried was LeNet made by Yan LeCun in 1998 with the following specification:

  1. The input is images 32x32x1
  2. Convolution layer 1. The output shape should be 28x28x6.
  3. Activation 1. ReLU function
  4. Pooling layer 1. The output shape is 14x14x6.
  5. Convolution layer 2. The output shape is 10x10x16.
  6. Activation 2. ReLU function
  7. Pooling layer 2. The output shape is 5x5x16.
  8. Flatten layer.
  9. Fully connected layer 1. Has 120 outputs.
  10. Activation 3. ReLU function
  11. Fully connected layer 2. Has 84 outputs.
  12. Activation 4. ReLU function
  13. Fully connected layer 3. Has 43 outputs.
In [353]:
### Train your model here.
### Feel free to use as many code cells as needed.
# Train the Model
with tf.Session() as sess:
    sess.run(tf.initialize_all_variables())
    num_examples = len(X_train_pp)
    
    print("Training...")
    print()
    for i in range(EPOCHS):
        X_train_pp, y_train = shuffle(X_train_pp, y_train)
        for offset in range(0, num_examples, BATCH_SIZE):
            end = offset + BATCH_SIZE
            batch_x, batch_y = X_train_pp[offset:end], y_train[offset:end]
            sess.run(training_operation, feed_dict={x: batch_x, y: batch_y})
            
        validation_accuracy = evaluate(X_validation, y_validation)
        print("EPOCH {} ...".format(i+1))
        print("Validation Accuracy = {:.3f}".format(validation_accuracy))
        print()
        
    saver.save(sess, 'lenet')
    print("Model saved")

Training...

EPOCH 1 ...
Validation Accuracy = 0.619

EPOCH 2 ...
Validation Accuracy = 0.817

EPOCH 3 ...
Validation Accuracy = 0.861

EPOCH 4 ...
Validation Accuracy = 0.903

EPOCH 5 ...
Validation Accuracy = 0.915

EPOCH 6 ...
Validation Accuracy = 0.924

EPOCH 7 ...
Validation Accuracy = 0.934

EPOCH 8 ...
Validation Accuracy = 0.936

EPOCH 9 ...
Validation Accuracy = 0.939

EPOCH 10 ...
Validation Accuracy = 0.938

Model saved
In [354]:
print ("Accuracy on test data")
with tf.Session() as sess:
    saver.restore(sess, tf.train.latest_checkpoint('.'))

    test_accuracy = evaluate(X_test_pp, y_test)
    print("Test Accuracy = {:.3f}".format(test_accuracy))

Accuracy on test data
Test Accuracy = 0.837

Question 4

How did you train your model? (Type of optimizer, batch size, epochs, hyperparameters, etc.)

Answer: In order to train and validate my model, I've done the following steps:

  1. Created the TensorFlow session and initiated all variables and trained the model over EPOCHs.
  2. At the beginning of each EPOCH I've shuffled the training data to ensure that our training is not biased by the order of the images in the training data set.
  3. Then I've broken the training data set into batches and train the model on each batch.
  4. At the end of each EPOCH I've validated the model on the validation data.
  5. Once the model is completely trained, it's saved.

Parameters:

  1. Optimizer: Adam algorithm to minimize the loss function.
  2. Batch size is 128.
  3. EPOCHs is 10.
  4. Learning rate is 0.001 (how quickly to update the network's weights).

Question 5

What approach did you take in coming up with a solution to this problem? It may have been a process of trial and error, in which case, outline the steps you took to get to the final solution and why you chose those steps. Perhaps your solution involved an already well known implementation or architecture. In this case, discuss why you think this is suitable for the current problem.

Answer: Current solution is a starting point and default choice. I've started from LeNet architecture as it proves the efficiency over the years and I had an experience with it on the MNiST data set. The model shows appropriate accuracy and might be a starting point for further experiments.
As a follow up steps I'm considering trying over well known implementations such as:

  1. AlexNet developed by Alex Krizhevsky, Ilya Sutskever and Geoff Hinton.
  2. ZF Net by Matthew Zeiler and Rob Fergus.
  3. GoogLeNet by Szegedy et al from Google.
  4. VGGNet by Karen Simonyan and Andrew Zisserman.
  5. ResNet (Residual Network) developed by Kaiming He et al.

http://cs231n.github.io/convolutional-networks/#case

Step 3: Test a Model on New Images

Take several pictures of traffic signs that you find on the web or around you (at least five), and run them through your classifier on your computer to produce example results. The classifier might not recognize some local signs but it could prove interesting nonetheless.

You may find signnames.csv useful as it contains mappings from the class id (integer) to the actual sign name.

In [359]:
import matplotlib
matplotlib.rcParams.update({'font.size': 22})

new_signs = 'traffic-signs-data/new_signs.p'

with open(new_signs, mode='rb') as f:
    test_my = pickle.load(f)
    
X_test_my = train_my['data']
y_test_my = np.array([28,2,22,22,36,
                            13,17,1,1,17,
                            25,38,25,31,35,
                            22,25,36,38,38,
                            17,22,35,12,33,
                            38,38,25,38
                           ],dtype=np.uint8)
    
X_test_my_pp = data_pp(X_test_my, gray=1, norm=1)
with tf.Session() as sess:
    saver.restore(sess, tf.train.latest_checkpoint('.'))
    test_accuracy = evaluate(X_test_my_pp, y_test_my)
    logits_ds = sess.run(logits, feed_dict={x: X_test_my_pp})
    
b = plt.figure(figsize=(32,32/5*29))

for i, image in enumerate(X_test_my):
    
    ax = b.add_subplot(len(X_test_my), 2, 2*i+1)
    plt.imshow(X_test_my[i])

    predict = (y_test_my[i] == np.argmax(logits_ds[i]))
    text = "Prediction: " + str(predict)
    ax.text(35, 2, text)
    text = "Given: " + signname(y_test_my[i])
    ax.text(35, 5, text)
    text = "Predicted: " + signname(np.argmax(logits_ds[i]))
    ax.text(35, 8, text)
    top5 = logits_ds[i][y_test_my[i]] in np.sort(logits_ds[i])[:37:-1]    
    text = "Prediction in TOP5: " + str(top5)
    ax.text(35, 11, text)

    ax = b.add_subplot(len(X_test_my), 2, 2*i+2)
    bi = plt.bar(range(43), logits_ds[i])
    bi[y_test_my[i]].set_color('g')
    
print("Test Accuracy = {:.3f}".format(test_accuracy))

Test Accuracy = 0.690
In [360]:
import os
import cv2

X_test_my = np.empty([5,32,32,3],dtype=np.uint8)
y_test_my = np.array([17, 0, 9, 9, 14])

test_images = os.listdir("my_signs/")
if '.DS_Store' in test_images:
    test_images.remove('.DS_Store')

for i, name in enumerate(test_images):
    X_test_my[i] = cv2.cvtColor(cv2.imread('my_signs/'+name), cv2.COLOR_BGR2RGB)

X_test_my_pp = data_pp(X_test_my, gray=1, norm=1)

with tf.Session() as sess:
    saver.restore(sess, tf.train.latest_checkpoint('.'))

    test_accuracy = evaluate(X_test_my_pp, y_test_my)
    logits_ds = sess.run(logits, feed_dict={x: X_test_my_pp})
    
b = plt.figure(figsize=(32,32))

for i, image in enumerate(X_test_my):
    
    ax = b.add_subplot(5, 2, 2*i+1)
    plt.imshow(X_test_my[i])

    predict = (y_test_my[i] == np.argmax(logits_ds[i]))
    text = "Prediction: " + str(predict)
    ax.text(35, 2, text)
    text = "Given: " + signname(y_test_my[i])
    ax.text(35, 5, text)
    text = "Predicted: " + signname(np.argmax(logits_ds[i]))
    ax.text(35, 8, text)
    top5 = logits_ds[i][y_test_my[i]] in np.sort(logits_ds[i])[:37:-1]    
    text = "Prediction in TOP5: " + str(top5)
    ax.text(35, 11, text)
    ax = b.add_subplot(5, 2, 2*i+2)
    bi = plt.bar(range(43), logits_ds[i])
    bi[y_test_my[i]].set_color('g')
    
print("Test Accuracy = {:.3f}".format(test_accuracy))

Test Accuracy = 0.200

Question 6

Choose five candidate images of traffic signs and provide them in the report. Are there any particular qualities of the image(s) that might make classification difficult? It could be helpful to plot the images in the notebook.

Answer: To test the model I've tried 2 datasets:

  1. First was gathered by Hanqiu Ying on the streets of Germany. All the signs were in the training data, quality is comparable to the training data.
  2. Second was just a collection of sign icons found in the internet. The signs looks visually different, zoomed in.

Question 7

Is your model able to perform equally well on captured pictures when compared to testing on the dataset? The simplest way to do this check the accuracy of the predictions. For example, if the model predicted 1 out of 5 signs correctly, it's 20% accurate.

NOTE: You could check the accuracy manually by using signnames.csv (same directory). This file has a mapping from the class id (0-42) to the corresponding sign name. So, you could take the class id the model outputs, lookup the name in signnames.csv and see if it matches the sign from the image.

Answer: First set showed comparable accuracy with the given data set - 0.828.
All 5 examples from the 2nd dataset have signs zoomed-in compared to the signs in the dataset and looks pretty different from the given signs.
Nevertheless, the model could recognize the signs (but the accuracy requires improvements). When the training dataset will be extended via image transformation in the future, I expect to get much better performance.

Question 8

Use the model's softmax probabilities to visualize the certainty of its predictions, tf.nn.top_k could prove helpful here. Which predictions is the model certain of? Uncertain? If the model was incorrect in its initial prediction, does the correct prediction appear in the top k? (k should be 5 at most)

tf.nn.top_k will return the values and indices (class ids) of the top k predictions. So if k=3, for each sign, it'll return the 3 largest probabilities (out of a possible 43) and the correspoding class ids.

Take this numpy array as an example:

# (5, 6) array
a = np.array([[ 0.24879643,  0.07032244,  0.12641572,  0.34763842,  0.07893497,
         0.12789202],
       [ 0.28086119,  0.27569815,  0.08594638,  0.0178669 ,  0.18063401,
         0.15899337],
       [ 0.26076848,  0.23664738,  0.08020603,  0.07001922,  0.1134371 ,
         0.23892179],
       [ 0.11943333,  0.29198961,  0.02605103,  0.26234032,  0.1351348 ,
         0.16505091],
       [ 0.09561176,  0.34396535,  0.0643941 ,  0.16240774,  0.24206137,
         0.09155967]])

Running it through sess.run(tf.nn.top_k(tf.constant(a), k=3)) produces:

TopKV2(values=array([[ 0.34763842,  0.24879643,  0.12789202],
       [ 0.28086119,  0.27569815,  0.18063401],
       [ 0.26076848,  0.23892179,  0.23664738],
       [ 0.29198961,  0.26234032,  0.16505091],
       [ 0.34396535,  0.24206137,  0.16240774]]), indices=array([[3, 0, 5],
       [0, 1, 4],
       [0, 5, 1],
       [1, 3, 5],
       [1, 4, 3]], dtype=int32))

Looking just at the first row we get [ 0.34763842, 0.24879643, 0.12789202], you can confirm these are the 3 largest probabilities in a. You'll also notice [3, 0, 5] are the corresponding indices.

Answer: I've added the feature "Prediction in TOP5" in the logits visualization section. In most of the cases if the prediction was incorrect, the correct class was also not among TOP5 predictions.
On the first dataset (real photos of germans' traffic signs) model made most of mistakes with blue signs. During the data preprocessing I've transformed images into grayscale which was probably not the most optimal solution as I've lost information about the color.
On the second dataset model was wrong on the signs which are zoomed in more than others. Data augmentation via rotation, shearing and translation should improve the current result.

Note: Once you have completed all of the code implementations and successfully answered each question above, you may finalize your work by exporting the iPython Notebook as an HTML document. You can do this by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

Resources

  1. Xiu-Shen Wei, Nanjing University. Must Know Tips for Deep Learning Neural Networks, Part 1. http://www.kdnuggets.com/2016/03/must-know-tips-deep-learning-part-1.html
  2. German sign classification using deep learning neural networks. Vivek Yadav. https://chatbotslife.com/german-sign-classification-using-deep-learning-neural-networks-98-8-solution-d05656bf51ad#.g7vfjpcoy
In [ ]: